FDE Implementation Workbench
The workbench models the implementation gap between a messy customer ask and a workflow that can survive validation, failures, policy checks, and human review.
Status
Public offline V1 · Synthetic fixtures
Capabilities
Implementation design · Approval workflows · Failure handling
The problem
Enterprise implementation work often fails after the demo because inputs are messy, tools fail, business rules conflict, and approval requirements are easy to bypass.
What I worked on
An eight-stage pipeline handles ingestion, normalization, tool execution, validation, policy, risk, approval, and final audit output. Mock adapters create deterministic failure cases for retry and escalation testing.
Decisions and tradeoffs
Treat messy inputs as expected data
Normalization records parse notes and routes recoverable issues through the workflow instead of crashing the run.
Bound every recovery path
Retryable failures have explicit attempt limits, and fatal failures do not enter a retry loop.
Make approval a hard invariant
Escalated items cannot reach the final payment step without an explicit approval identity and reason.
Evidence and scope
The repository uses synthetic fixtures and mock external systems. It demonstrates implementation patterns and does not claim production customer deployment.
What this shows
Customer-workflow thinking, error taxonomy, implementation discipline, and the ability to translate ambiguous requirements into an auditable operating system.
Implementation details
Key subsystems
- Eight-stage workflow pipeline
- Four mock tool adapters
- Bounded retry and escalation logic
- CLI, local HTTP API, and JSONL audit trail